xref: /openbmc/linux/fs/ubifs/lpt_commit.c (revision 4bce6fce)
1 /*
2  * This file is part of UBIFS.
3  *
4  * Copyright (C) 2006-2008 Nokia Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published by
8  * the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program; if not, write to the Free Software Foundation, Inc., 51
17  * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18  *
19  * Authors: Adrian Hunter
20  *          Artem Bityutskiy (Битюцкий Артём)
21  */
22 
23 /*
24  * This file implements commit-related functionality of the LEB properties
25  * subsystem.
26  */
27 
28 #include <linux/crc16.h>
29 #include <linux/slab.h>
30 #include <linux/random.h>
31 #include "ubifs.h"
32 
33 static int dbg_populate_lsave(struct ubifs_info *c);
34 
35 /**
36  * first_dirty_cnode - find first dirty cnode.
37  * @c: UBIFS file-system description object
38  * @nnode: nnode at which to start
39  *
40  * This function returns the first dirty cnode or %NULL if there is not one.
41  */
42 static struct ubifs_cnode *first_dirty_cnode(struct ubifs_nnode *nnode)
43 {
44 	ubifs_assert(nnode);
45 	while (1) {
46 		int i, cont = 0;
47 
48 		for (i = 0; i < UBIFS_LPT_FANOUT; i++) {
49 			struct ubifs_cnode *cnode;
50 
51 			cnode = nnode->nbranch[i].cnode;
52 			if (cnode &&
53 			    test_bit(DIRTY_CNODE, &cnode->flags)) {
54 				if (cnode->level == 0)
55 					return cnode;
56 				nnode = (struct ubifs_nnode *)cnode;
57 				cont = 1;
58 				break;
59 			}
60 		}
61 		if (!cont)
62 			return (struct ubifs_cnode *)nnode;
63 	}
64 }
65 
66 /**
67  * next_dirty_cnode - find next dirty cnode.
68  * @cnode: cnode from which to begin searching
69  *
70  * This function returns the next dirty cnode or %NULL if there is not one.
71  */
72 static struct ubifs_cnode *next_dirty_cnode(struct ubifs_cnode *cnode)
73 {
74 	struct ubifs_nnode *nnode;
75 	int i;
76 
77 	ubifs_assert(cnode);
78 	nnode = cnode->parent;
79 	if (!nnode)
80 		return NULL;
81 	for (i = cnode->iip + 1; i < UBIFS_LPT_FANOUT; i++) {
82 		cnode = nnode->nbranch[i].cnode;
83 		if (cnode && test_bit(DIRTY_CNODE, &cnode->flags)) {
84 			if (cnode->level == 0)
85 				return cnode; /* cnode is a pnode */
86 			/* cnode is a nnode */
87 			return first_dirty_cnode((struct ubifs_nnode *)cnode);
88 		}
89 	}
90 	return (struct ubifs_cnode *)nnode;
91 }
92 
93 /**
94  * get_cnodes_to_commit - create list of dirty cnodes to commit.
95  * @c: UBIFS file-system description object
96  *
97  * This function returns the number of cnodes to commit.
98  */
99 static int get_cnodes_to_commit(struct ubifs_info *c)
100 {
101 	struct ubifs_cnode *cnode, *cnext;
102 	int cnt = 0;
103 
104 	if (!c->nroot)
105 		return 0;
106 
107 	if (!test_bit(DIRTY_CNODE, &c->nroot->flags))
108 		return 0;
109 
110 	c->lpt_cnext = first_dirty_cnode(c->nroot);
111 	cnode = c->lpt_cnext;
112 	if (!cnode)
113 		return 0;
114 	cnt += 1;
115 	while (1) {
116 		ubifs_assert(!test_bit(COW_CNODE, &cnode->flags));
117 		__set_bit(COW_CNODE, &cnode->flags);
118 		cnext = next_dirty_cnode(cnode);
119 		if (!cnext) {
120 			cnode->cnext = c->lpt_cnext;
121 			break;
122 		}
123 		cnode->cnext = cnext;
124 		cnode = cnext;
125 		cnt += 1;
126 	}
127 	dbg_cmt("committing %d cnodes", cnt);
128 	dbg_lp("committing %d cnodes", cnt);
129 	ubifs_assert(cnt == c->dirty_nn_cnt + c->dirty_pn_cnt);
130 	return cnt;
131 }
132 
133 /**
134  * upd_ltab - update LPT LEB properties.
135  * @c: UBIFS file-system description object
136  * @lnum: LEB number
137  * @free: amount of free space
138  * @dirty: amount of dirty space to add
139  */
140 static void upd_ltab(struct ubifs_info *c, int lnum, int free, int dirty)
141 {
142 	dbg_lp("LEB %d free %d dirty %d to %d +%d",
143 	       lnum, c->ltab[lnum - c->lpt_first].free,
144 	       c->ltab[lnum - c->lpt_first].dirty, free, dirty);
145 	ubifs_assert(lnum >= c->lpt_first && lnum <= c->lpt_last);
146 	c->ltab[lnum - c->lpt_first].free = free;
147 	c->ltab[lnum - c->lpt_first].dirty += dirty;
148 }
149 
150 /**
151  * alloc_lpt_leb - allocate an LPT LEB that is empty.
152  * @c: UBIFS file-system description object
153  * @lnum: LEB number is passed and returned here
154  *
155  * This function finds the next empty LEB in the ltab starting from @lnum. If a
156  * an empty LEB is found it is returned in @lnum and the function returns %0.
157  * Otherwise the function returns -ENOSPC.  Note however, that LPT is designed
158  * never to run out of space.
159  */
160 static int alloc_lpt_leb(struct ubifs_info *c, int *lnum)
161 {
162 	int i, n;
163 
164 	n = *lnum - c->lpt_first + 1;
165 	for (i = n; i < c->lpt_lebs; i++) {
166 		if (c->ltab[i].tgc || c->ltab[i].cmt)
167 			continue;
168 		if (c->ltab[i].free == c->leb_size) {
169 			c->ltab[i].cmt = 1;
170 			*lnum = i + c->lpt_first;
171 			return 0;
172 		}
173 	}
174 
175 	for (i = 0; i < n; i++) {
176 		if (c->ltab[i].tgc || c->ltab[i].cmt)
177 			continue;
178 		if (c->ltab[i].free == c->leb_size) {
179 			c->ltab[i].cmt = 1;
180 			*lnum = i + c->lpt_first;
181 			return 0;
182 		}
183 	}
184 	return -ENOSPC;
185 }
186 
187 /**
188  * layout_cnodes - layout cnodes for commit.
189  * @c: UBIFS file-system description object
190  *
191  * This function returns %0 on success and a negative error code on failure.
192  */
193 static int layout_cnodes(struct ubifs_info *c)
194 {
195 	int lnum, offs, len, alen, done_lsave, done_ltab, err;
196 	struct ubifs_cnode *cnode;
197 
198 	err = dbg_chk_lpt_sz(c, 0, 0);
199 	if (err)
200 		return err;
201 	cnode = c->lpt_cnext;
202 	if (!cnode)
203 		return 0;
204 	lnum = c->nhead_lnum;
205 	offs = c->nhead_offs;
206 	/* Try to place lsave and ltab nicely */
207 	done_lsave = !c->big_lpt;
208 	done_ltab = 0;
209 	if (!done_lsave && offs + c->lsave_sz <= c->leb_size) {
210 		done_lsave = 1;
211 		c->lsave_lnum = lnum;
212 		c->lsave_offs = offs;
213 		offs += c->lsave_sz;
214 		dbg_chk_lpt_sz(c, 1, c->lsave_sz);
215 	}
216 
217 	if (offs + c->ltab_sz <= c->leb_size) {
218 		done_ltab = 1;
219 		c->ltab_lnum = lnum;
220 		c->ltab_offs = offs;
221 		offs += c->ltab_sz;
222 		dbg_chk_lpt_sz(c, 1, c->ltab_sz);
223 	}
224 
225 	do {
226 		if (cnode->level) {
227 			len = c->nnode_sz;
228 			c->dirty_nn_cnt -= 1;
229 		} else {
230 			len = c->pnode_sz;
231 			c->dirty_pn_cnt -= 1;
232 		}
233 		while (offs + len > c->leb_size) {
234 			alen = ALIGN(offs, c->min_io_size);
235 			upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
236 			dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
237 			err = alloc_lpt_leb(c, &lnum);
238 			if (err)
239 				goto no_space;
240 			offs = 0;
241 			ubifs_assert(lnum >= c->lpt_first &&
242 				     lnum <= c->lpt_last);
243 			/* Try to place lsave and ltab nicely */
244 			if (!done_lsave) {
245 				done_lsave = 1;
246 				c->lsave_lnum = lnum;
247 				c->lsave_offs = offs;
248 				offs += c->lsave_sz;
249 				dbg_chk_lpt_sz(c, 1, c->lsave_sz);
250 				continue;
251 			}
252 			if (!done_ltab) {
253 				done_ltab = 1;
254 				c->ltab_lnum = lnum;
255 				c->ltab_offs = offs;
256 				offs += c->ltab_sz;
257 				dbg_chk_lpt_sz(c, 1, c->ltab_sz);
258 				continue;
259 			}
260 			break;
261 		}
262 		if (cnode->parent) {
263 			cnode->parent->nbranch[cnode->iip].lnum = lnum;
264 			cnode->parent->nbranch[cnode->iip].offs = offs;
265 		} else {
266 			c->lpt_lnum = lnum;
267 			c->lpt_offs = offs;
268 		}
269 		offs += len;
270 		dbg_chk_lpt_sz(c, 1, len);
271 		cnode = cnode->cnext;
272 	} while (cnode && cnode != c->lpt_cnext);
273 
274 	/* Make sure to place LPT's save table */
275 	if (!done_lsave) {
276 		if (offs + c->lsave_sz > c->leb_size) {
277 			alen = ALIGN(offs, c->min_io_size);
278 			upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
279 			dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
280 			err = alloc_lpt_leb(c, &lnum);
281 			if (err)
282 				goto no_space;
283 			offs = 0;
284 			ubifs_assert(lnum >= c->lpt_first &&
285 				     lnum <= c->lpt_last);
286 		}
287 		done_lsave = 1;
288 		c->lsave_lnum = lnum;
289 		c->lsave_offs = offs;
290 		offs += c->lsave_sz;
291 		dbg_chk_lpt_sz(c, 1, c->lsave_sz);
292 	}
293 
294 	/* Make sure to place LPT's own lprops table */
295 	if (!done_ltab) {
296 		if (offs + c->ltab_sz > c->leb_size) {
297 			alen = ALIGN(offs, c->min_io_size);
298 			upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
299 			dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
300 			err = alloc_lpt_leb(c, &lnum);
301 			if (err)
302 				goto no_space;
303 			offs = 0;
304 			ubifs_assert(lnum >= c->lpt_first &&
305 				     lnum <= c->lpt_last);
306 		}
307 		c->ltab_lnum = lnum;
308 		c->ltab_offs = offs;
309 		offs += c->ltab_sz;
310 		dbg_chk_lpt_sz(c, 1, c->ltab_sz);
311 	}
312 
313 	alen = ALIGN(offs, c->min_io_size);
314 	upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
315 	dbg_chk_lpt_sz(c, 4, alen - offs);
316 	err = dbg_chk_lpt_sz(c, 3, alen);
317 	if (err)
318 		return err;
319 	return 0;
320 
321 no_space:
322 	ubifs_err(c, "LPT out of space at LEB %d:%d needing %d, done_ltab %d, done_lsave %d",
323 		  lnum, offs, len, done_ltab, done_lsave);
324 	ubifs_dump_lpt_info(c);
325 	ubifs_dump_lpt_lebs(c);
326 	dump_stack();
327 	return err;
328 }
329 
330 /**
331  * realloc_lpt_leb - allocate an LPT LEB that is empty.
332  * @c: UBIFS file-system description object
333  * @lnum: LEB number is passed and returned here
334  *
335  * This function duplicates exactly the results of the function alloc_lpt_leb.
336  * It is used during end commit to reallocate the same LEB numbers that were
337  * allocated by alloc_lpt_leb during start commit.
338  *
339  * This function finds the next LEB that was allocated by the alloc_lpt_leb
340  * function starting from @lnum. If a LEB is found it is returned in @lnum and
341  * the function returns %0. Otherwise the function returns -ENOSPC.
342  * Note however, that LPT is designed never to run out of space.
343  */
344 static int realloc_lpt_leb(struct ubifs_info *c, int *lnum)
345 {
346 	int i, n;
347 
348 	n = *lnum - c->lpt_first + 1;
349 	for (i = n; i < c->lpt_lebs; i++)
350 		if (c->ltab[i].cmt) {
351 			c->ltab[i].cmt = 0;
352 			*lnum = i + c->lpt_first;
353 			return 0;
354 		}
355 
356 	for (i = 0; i < n; i++)
357 		if (c->ltab[i].cmt) {
358 			c->ltab[i].cmt = 0;
359 			*lnum = i + c->lpt_first;
360 			return 0;
361 		}
362 	return -ENOSPC;
363 }
364 
365 /**
366  * write_cnodes - write cnodes for commit.
367  * @c: UBIFS file-system description object
368  *
369  * This function returns %0 on success and a negative error code on failure.
370  */
371 static int write_cnodes(struct ubifs_info *c)
372 {
373 	int lnum, offs, len, from, err, wlen, alen, done_ltab, done_lsave;
374 	struct ubifs_cnode *cnode;
375 	void *buf = c->lpt_buf;
376 
377 	cnode = c->lpt_cnext;
378 	if (!cnode)
379 		return 0;
380 	lnum = c->nhead_lnum;
381 	offs = c->nhead_offs;
382 	from = offs;
383 	/* Ensure empty LEB is unmapped */
384 	if (offs == 0) {
385 		err = ubifs_leb_unmap(c, lnum);
386 		if (err)
387 			return err;
388 	}
389 	/* Try to place lsave and ltab nicely */
390 	done_lsave = !c->big_lpt;
391 	done_ltab = 0;
392 	if (!done_lsave && offs + c->lsave_sz <= c->leb_size) {
393 		done_lsave = 1;
394 		ubifs_pack_lsave(c, buf + offs, c->lsave);
395 		offs += c->lsave_sz;
396 		dbg_chk_lpt_sz(c, 1, c->lsave_sz);
397 	}
398 
399 	if (offs + c->ltab_sz <= c->leb_size) {
400 		done_ltab = 1;
401 		ubifs_pack_ltab(c, buf + offs, c->ltab_cmt);
402 		offs += c->ltab_sz;
403 		dbg_chk_lpt_sz(c, 1, c->ltab_sz);
404 	}
405 
406 	/* Loop for each cnode */
407 	do {
408 		if (cnode->level)
409 			len = c->nnode_sz;
410 		else
411 			len = c->pnode_sz;
412 		while (offs + len > c->leb_size) {
413 			wlen = offs - from;
414 			if (wlen) {
415 				alen = ALIGN(wlen, c->min_io_size);
416 				memset(buf + offs, 0xff, alen - wlen);
417 				err = ubifs_leb_write(c, lnum, buf + from, from,
418 						       alen);
419 				if (err)
420 					return err;
421 			}
422 			dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
423 			err = realloc_lpt_leb(c, &lnum);
424 			if (err)
425 				goto no_space;
426 			offs = from = 0;
427 			ubifs_assert(lnum >= c->lpt_first &&
428 				     lnum <= c->lpt_last);
429 			err = ubifs_leb_unmap(c, lnum);
430 			if (err)
431 				return err;
432 			/* Try to place lsave and ltab nicely */
433 			if (!done_lsave) {
434 				done_lsave = 1;
435 				ubifs_pack_lsave(c, buf + offs, c->lsave);
436 				offs += c->lsave_sz;
437 				dbg_chk_lpt_sz(c, 1, c->lsave_sz);
438 				continue;
439 			}
440 			if (!done_ltab) {
441 				done_ltab = 1;
442 				ubifs_pack_ltab(c, buf + offs, c->ltab_cmt);
443 				offs += c->ltab_sz;
444 				dbg_chk_lpt_sz(c, 1, c->ltab_sz);
445 				continue;
446 			}
447 			break;
448 		}
449 		if (cnode->level)
450 			ubifs_pack_nnode(c, buf + offs,
451 					 (struct ubifs_nnode *)cnode);
452 		else
453 			ubifs_pack_pnode(c, buf + offs,
454 					 (struct ubifs_pnode *)cnode);
455 		/*
456 		 * The reason for the barriers is the same as in case of TNC.
457 		 * See comment in 'write_index()'. 'dirty_cow_nnode()' and
458 		 * 'dirty_cow_pnode()' are the functions for which this is
459 		 * important.
460 		 */
461 		clear_bit(DIRTY_CNODE, &cnode->flags);
462 		smp_mb__before_atomic();
463 		clear_bit(COW_CNODE, &cnode->flags);
464 		smp_mb__after_atomic();
465 		offs += len;
466 		dbg_chk_lpt_sz(c, 1, len);
467 		cnode = cnode->cnext;
468 	} while (cnode && cnode != c->lpt_cnext);
469 
470 	/* Make sure to place LPT's save table */
471 	if (!done_lsave) {
472 		if (offs + c->lsave_sz > c->leb_size) {
473 			wlen = offs - from;
474 			alen = ALIGN(wlen, c->min_io_size);
475 			memset(buf + offs, 0xff, alen - wlen);
476 			err = ubifs_leb_write(c, lnum, buf + from, from, alen);
477 			if (err)
478 				return err;
479 			dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
480 			err = realloc_lpt_leb(c, &lnum);
481 			if (err)
482 				goto no_space;
483 			offs = from = 0;
484 			ubifs_assert(lnum >= c->lpt_first &&
485 				     lnum <= c->lpt_last);
486 			err = ubifs_leb_unmap(c, lnum);
487 			if (err)
488 				return err;
489 		}
490 		done_lsave = 1;
491 		ubifs_pack_lsave(c, buf + offs, c->lsave);
492 		offs += c->lsave_sz;
493 		dbg_chk_lpt_sz(c, 1, c->lsave_sz);
494 	}
495 
496 	/* Make sure to place LPT's own lprops table */
497 	if (!done_ltab) {
498 		if (offs + c->ltab_sz > c->leb_size) {
499 			wlen = offs - from;
500 			alen = ALIGN(wlen, c->min_io_size);
501 			memset(buf + offs, 0xff, alen - wlen);
502 			err = ubifs_leb_write(c, lnum, buf + from, from, alen);
503 			if (err)
504 				return err;
505 			dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
506 			err = realloc_lpt_leb(c, &lnum);
507 			if (err)
508 				goto no_space;
509 			offs = from = 0;
510 			ubifs_assert(lnum >= c->lpt_first &&
511 				     lnum <= c->lpt_last);
512 			err = ubifs_leb_unmap(c, lnum);
513 			if (err)
514 				return err;
515 		}
516 		ubifs_pack_ltab(c, buf + offs, c->ltab_cmt);
517 		offs += c->ltab_sz;
518 		dbg_chk_lpt_sz(c, 1, c->ltab_sz);
519 	}
520 
521 	/* Write remaining data in buffer */
522 	wlen = offs - from;
523 	alen = ALIGN(wlen, c->min_io_size);
524 	memset(buf + offs, 0xff, alen - wlen);
525 	err = ubifs_leb_write(c, lnum, buf + from, from, alen);
526 	if (err)
527 		return err;
528 
529 	dbg_chk_lpt_sz(c, 4, alen - wlen);
530 	err = dbg_chk_lpt_sz(c, 3, ALIGN(offs, c->min_io_size));
531 	if (err)
532 		return err;
533 
534 	c->nhead_lnum = lnum;
535 	c->nhead_offs = ALIGN(offs, c->min_io_size);
536 
537 	dbg_lp("LPT root is at %d:%d", c->lpt_lnum, c->lpt_offs);
538 	dbg_lp("LPT head is at %d:%d", c->nhead_lnum, c->nhead_offs);
539 	dbg_lp("LPT ltab is at %d:%d", c->ltab_lnum, c->ltab_offs);
540 	if (c->big_lpt)
541 		dbg_lp("LPT lsave is at %d:%d", c->lsave_lnum, c->lsave_offs);
542 
543 	return 0;
544 
545 no_space:
546 	ubifs_err(c, "LPT out of space mismatch at LEB %d:%d needing %d, done_ltab %d, done_lsave %d",
547 		  lnum, offs, len, done_ltab, done_lsave);
548 	ubifs_dump_lpt_info(c);
549 	ubifs_dump_lpt_lebs(c);
550 	dump_stack();
551 	return err;
552 }
553 
554 /**
555  * next_pnode_to_dirty - find next pnode to dirty.
556  * @c: UBIFS file-system description object
557  * @pnode: pnode
558  *
559  * This function returns the next pnode to dirty or %NULL if there are no more
560  * pnodes.  Note that pnodes that have never been written (lnum == 0) are
561  * skipped.
562  */
563 static struct ubifs_pnode *next_pnode_to_dirty(struct ubifs_info *c,
564 					       struct ubifs_pnode *pnode)
565 {
566 	struct ubifs_nnode *nnode;
567 	int iip;
568 
569 	/* Try to go right */
570 	nnode = pnode->parent;
571 	for (iip = pnode->iip + 1; iip < UBIFS_LPT_FANOUT; iip++) {
572 		if (nnode->nbranch[iip].lnum)
573 			return ubifs_get_pnode(c, nnode, iip);
574 	}
575 
576 	/* Go up while can't go right */
577 	do {
578 		iip = nnode->iip + 1;
579 		nnode = nnode->parent;
580 		if (!nnode)
581 			return NULL;
582 		for (; iip < UBIFS_LPT_FANOUT; iip++) {
583 			if (nnode->nbranch[iip].lnum)
584 				break;
585 		}
586 	} while (iip >= UBIFS_LPT_FANOUT);
587 
588 	/* Go right */
589 	nnode = ubifs_get_nnode(c, nnode, iip);
590 	if (IS_ERR(nnode))
591 		return (void *)nnode;
592 
593 	/* Go down to level 1 */
594 	while (nnode->level > 1) {
595 		for (iip = 0; iip < UBIFS_LPT_FANOUT; iip++) {
596 			if (nnode->nbranch[iip].lnum)
597 				break;
598 		}
599 		if (iip >= UBIFS_LPT_FANOUT) {
600 			/*
601 			 * Should not happen, but we need to keep going
602 			 * if it does.
603 			 */
604 			iip = 0;
605 		}
606 		nnode = ubifs_get_nnode(c, nnode, iip);
607 		if (IS_ERR(nnode))
608 			return (void *)nnode;
609 	}
610 
611 	for (iip = 0; iip < UBIFS_LPT_FANOUT; iip++)
612 		if (nnode->nbranch[iip].lnum)
613 			break;
614 	if (iip >= UBIFS_LPT_FANOUT)
615 		/* Should not happen, but we need to keep going if it does */
616 		iip = 0;
617 	return ubifs_get_pnode(c, nnode, iip);
618 }
619 
620 /**
621  * pnode_lookup - lookup a pnode in the LPT.
622  * @c: UBIFS file-system description object
623  * @i: pnode number (0 to main_lebs - 1)
624  *
625  * This function returns a pointer to the pnode on success or a negative
626  * error code on failure.
627  */
628 static struct ubifs_pnode *pnode_lookup(struct ubifs_info *c, int i)
629 {
630 	int err, h, iip, shft;
631 	struct ubifs_nnode *nnode;
632 
633 	if (!c->nroot) {
634 		err = ubifs_read_nnode(c, NULL, 0);
635 		if (err)
636 			return ERR_PTR(err);
637 	}
638 	i <<= UBIFS_LPT_FANOUT_SHIFT;
639 	nnode = c->nroot;
640 	shft = c->lpt_hght * UBIFS_LPT_FANOUT_SHIFT;
641 	for (h = 1; h < c->lpt_hght; h++) {
642 		iip = ((i >> shft) & (UBIFS_LPT_FANOUT - 1));
643 		shft -= UBIFS_LPT_FANOUT_SHIFT;
644 		nnode = ubifs_get_nnode(c, nnode, iip);
645 		if (IS_ERR(nnode))
646 			return ERR_CAST(nnode);
647 	}
648 	iip = ((i >> shft) & (UBIFS_LPT_FANOUT - 1));
649 	return ubifs_get_pnode(c, nnode, iip);
650 }
651 
652 /**
653  * add_pnode_dirt - add dirty space to LPT LEB properties.
654  * @c: UBIFS file-system description object
655  * @pnode: pnode for which to add dirt
656  */
657 static void add_pnode_dirt(struct ubifs_info *c, struct ubifs_pnode *pnode)
658 {
659 	ubifs_add_lpt_dirt(c, pnode->parent->nbranch[pnode->iip].lnum,
660 			   c->pnode_sz);
661 }
662 
663 /**
664  * do_make_pnode_dirty - mark a pnode dirty.
665  * @c: UBIFS file-system description object
666  * @pnode: pnode to mark dirty
667  */
668 static void do_make_pnode_dirty(struct ubifs_info *c, struct ubifs_pnode *pnode)
669 {
670 	/* Assumes cnext list is empty i.e. not called during commit */
671 	if (!test_and_set_bit(DIRTY_CNODE, &pnode->flags)) {
672 		struct ubifs_nnode *nnode;
673 
674 		c->dirty_pn_cnt += 1;
675 		add_pnode_dirt(c, pnode);
676 		/* Mark parent and ancestors dirty too */
677 		nnode = pnode->parent;
678 		while (nnode) {
679 			if (!test_and_set_bit(DIRTY_CNODE, &nnode->flags)) {
680 				c->dirty_nn_cnt += 1;
681 				ubifs_add_nnode_dirt(c, nnode);
682 				nnode = nnode->parent;
683 			} else
684 				break;
685 		}
686 	}
687 }
688 
689 /**
690  * make_tree_dirty - mark the entire LEB properties tree dirty.
691  * @c: UBIFS file-system description object
692  *
693  * This function is used by the "small" LPT model to cause the entire LEB
694  * properties tree to be written.  The "small" LPT model does not use LPT
695  * garbage collection because it is more efficient to write the entire tree
696  * (because it is small).
697  *
698  * This function returns %0 on success and a negative error code on failure.
699  */
700 static int make_tree_dirty(struct ubifs_info *c)
701 {
702 	struct ubifs_pnode *pnode;
703 
704 	pnode = pnode_lookup(c, 0);
705 	if (IS_ERR(pnode))
706 		return PTR_ERR(pnode);
707 
708 	while (pnode) {
709 		do_make_pnode_dirty(c, pnode);
710 		pnode = next_pnode_to_dirty(c, pnode);
711 		if (IS_ERR(pnode))
712 			return PTR_ERR(pnode);
713 	}
714 	return 0;
715 }
716 
717 /**
718  * need_write_all - determine if the LPT area is running out of free space.
719  * @c: UBIFS file-system description object
720  *
721  * This function returns %1 if the LPT area is running out of free space and %0
722  * if it is not.
723  */
724 static int need_write_all(struct ubifs_info *c)
725 {
726 	long long free = 0;
727 	int i;
728 
729 	for (i = 0; i < c->lpt_lebs; i++) {
730 		if (i + c->lpt_first == c->nhead_lnum)
731 			free += c->leb_size - c->nhead_offs;
732 		else if (c->ltab[i].free == c->leb_size)
733 			free += c->leb_size;
734 		else if (c->ltab[i].free + c->ltab[i].dirty == c->leb_size)
735 			free += c->leb_size;
736 	}
737 	/* Less than twice the size left */
738 	if (free <= c->lpt_sz * 2)
739 		return 1;
740 	return 0;
741 }
742 
743 /**
744  * lpt_tgc_start - start trivial garbage collection of LPT LEBs.
745  * @c: UBIFS file-system description object
746  *
747  * LPT trivial garbage collection is where a LPT LEB contains only dirty and
748  * free space and so may be reused as soon as the next commit is completed.
749  * This function is called during start commit to mark LPT LEBs for trivial GC.
750  */
751 static void lpt_tgc_start(struct ubifs_info *c)
752 {
753 	int i;
754 
755 	for (i = 0; i < c->lpt_lebs; i++) {
756 		if (i + c->lpt_first == c->nhead_lnum)
757 			continue;
758 		if (c->ltab[i].dirty > 0 &&
759 		    c->ltab[i].free + c->ltab[i].dirty == c->leb_size) {
760 			c->ltab[i].tgc = 1;
761 			c->ltab[i].free = c->leb_size;
762 			c->ltab[i].dirty = 0;
763 			dbg_lp("LEB %d", i + c->lpt_first);
764 		}
765 	}
766 }
767 
768 /**
769  * lpt_tgc_end - end trivial garbage collection of LPT LEBs.
770  * @c: UBIFS file-system description object
771  *
772  * LPT trivial garbage collection is where a LPT LEB contains only dirty and
773  * free space and so may be reused as soon as the next commit is completed.
774  * This function is called after the commit is completed (master node has been
775  * written) and un-maps LPT LEBs that were marked for trivial GC.
776  */
777 static int lpt_tgc_end(struct ubifs_info *c)
778 {
779 	int i, err;
780 
781 	for (i = 0; i < c->lpt_lebs; i++)
782 		if (c->ltab[i].tgc) {
783 			err = ubifs_leb_unmap(c, i + c->lpt_first);
784 			if (err)
785 				return err;
786 			c->ltab[i].tgc = 0;
787 			dbg_lp("LEB %d", i + c->lpt_first);
788 		}
789 	return 0;
790 }
791 
792 /**
793  * populate_lsave - fill the lsave array with important LEB numbers.
794  * @c: the UBIFS file-system description object
795  *
796  * This function is only called for the "big" model. It records a small number
797  * of LEB numbers of important LEBs.  Important LEBs are ones that are (from
798  * most important to least important): empty, freeable, freeable index, dirty
799  * index, dirty or free. Upon mount, we read this list of LEB numbers and bring
800  * their pnodes into memory.  That will stop us from having to scan the LPT
801  * straight away. For the "small" model we assume that scanning the LPT is no
802  * big deal.
803  */
804 static void populate_lsave(struct ubifs_info *c)
805 {
806 	struct ubifs_lprops *lprops;
807 	struct ubifs_lpt_heap *heap;
808 	int i, cnt = 0;
809 
810 	ubifs_assert(c->big_lpt);
811 	if (!(c->lpt_drty_flgs & LSAVE_DIRTY)) {
812 		c->lpt_drty_flgs |= LSAVE_DIRTY;
813 		ubifs_add_lpt_dirt(c, c->lsave_lnum, c->lsave_sz);
814 	}
815 
816 	if (dbg_populate_lsave(c))
817 		return;
818 
819 	list_for_each_entry(lprops, &c->empty_list, list) {
820 		c->lsave[cnt++] = lprops->lnum;
821 		if (cnt >= c->lsave_cnt)
822 			return;
823 	}
824 	list_for_each_entry(lprops, &c->freeable_list, list) {
825 		c->lsave[cnt++] = lprops->lnum;
826 		if (cnt >= c->lsave_cnt)
827 			return;
828 	}
829 	list_for_each_entry(lprops, &c->frdi_idx_list, list) {
830 		c->lsave[cnt++] = lprops->lnum;
831 		if (cnt >= c->lsave_cnt)
832 			return;
833 	}
834 	heap = &c->lpt_heap[LPROPS_DIRTY_IDX - 1];
835 	for (i = 0; i < heap->cnt; i++) {
836 		c->lsave[cnt++] = heap->arr[i]->lnum;
837 		if (cnt >= c->lsave_cnt)
838 			return;
839 	}
840 	heap = &c->lpt_heap[LPROPS_DIRTY - 1];
841 	for (i = 0; i < heap->cnt; i++) {
842 		c->lsave[cnt++] = heap->arr[i]->lnum;
843 		if (cnt >= c->lsave_cnt)
844 			return;
845 	}
846 	heap = &c->lpt_heap[LPROPS_FREE - 1];
847 	for (i = 0; i < heap->cnt; i++) {
848 		c->lsave[cnt++] = heap->arr[i]->lnum;
849 		if (cnt >= c->lsave_cnt)
850 			return;
851 	}
852 	/* Fill it up completely */
853 	while (cnt < c->lsave_cnt)
854 		c->lsave[cnt++] = c->main_first;
855 }
856 
857 /**
858  * nnode_lookup - lookup a nnode in the LPT.
859  * @c: UBIFS file-system description object
860  * @i: nnode number
861  *
862  * This function returns a pointer to the nnode on success or a negative
863  * error code on failure.
864  */
865 static struct ubifs_nnode *nnode_lookup(struct ubifs_info *c, int i)
866 {
867 	int err, iip;
868 	struct ubifs_nnode *nnode;
869 
870 	if (!c->nroot) {
871 		err = ubifs_read_nnode(c, NULL, 0);
872 		if (err)
873 			return ERR_PTR(err);
874 	}
875 	nnode = c->nroot;
876 	while (1) {
877 		iip = i & (UBIFS_LPT_FANOUT - 1);
878 		i >>= UBIFS_LPT_FANOUT_SHIFT;
879 		if (!i)
880 			break;
881 		nnode = ubifs_get_nnode(c, nnode, iip);
882 		if (IS_ERR(nnode))
883 			return nnode;
884 	}
885 	return nnode;
886 }
887 
888 /**
889  * make_nnode_dirty - find a nnode and, if found, make it dirty.
890  * @c: UBIFS file-system description object
891  * @node_num: nnode number of nnode to make dirty
892  * @lnum: LEB number where nnode was written
893  * @offs: offset where nnode was written
894  *
895  * This function is used by LPT garbage collection.  LPT garbage collection is
896  * used only for the "big" LPT model (c->big_lpt == 1).  Garbage collection
897  * simply involves marking all the nodes in the LEB being garbage-collected as
898  * dirty.  The dirty nodes are written next commit, after which the LEB is free
899  * to be reused.
900  *
901  * This function returns %0 on success and a negative error code on failure.
902  */
903 static int make_nnode_dirty(struct ubifs_info *c, int node_num, int lnum,
904 			    int offs)
905 {
906 	struct ubifs_nnode *nnode;
907 
908 	nnode = nnode_lookup(c, node_num);
909 	if (IS_ERR(nnode))
910 		return PTR_ERR(nnode);
911 	if (nnode->parent) {
912 		struct ubifs_nbranch *branch;
913 
914 		branch = &nnode->parent->nbranch[nnode->iip];
915 		if (branch->lnum != lnum || branch->offs != offs)
916 			return 0; /* nnode is obsolete */
917 	} else if (c->lpt_lnum != lnum || c->lpt_offs != offs)
918 			return 0; /* nnode is obsolete */
919 	/* Assumes cnext list is empty i.e. not called during commit */
920 	if (!test_and_set_bit(DIRTY_CNODE, &nnode->flags)) {
921 		c->dirty_nn_cnt += 1;
922 		ubifs_add_nnode_dirt(c, nnode);
923 		/* Mark parent and ancestors dirty too */
924 		nnode = nnode->parent;
925 		while (nnode) {
926 			if (!test_and_set_bit(DIRTY_CNODE, &nnode->flags)) {
927 				c->dirty_nn_cnt += 1;
928 				ubifs_add_nnode_dirt(c, nnode);
929 				nnode = nnode->parent;
930 			} else
931 				break;
932 		}
933 	}
934 	return 0;
935 }
936 
937 /**
938  * make_pnode_dirty - find a pnode and, if found, make it dirty.
939  * @c: UBIFS file-system description object
940  * @node_num: pnode number of pnode to make dirty
941  * @lnum: LEB number where pnode was written
942  * @offs: offset where pnode was written
943  *
944  * This function is used by LPT garbage collection.  LPT garbage collection is
945  * used only for the "big" LPT model (c->big_lpt == 1).  Garbage collection
946  * simply involves marking all the nodes in the LEB being garbage-collected as
947  * dirty.  The dirty nodes are written next commit, after which the LEB is free
948  * to be reused.
949  *
950  * This function returns %0 on success and a negative error code on failure.
951  */
952 static int make_pnode_dirty(struct ubifs_info *c, int node_num, int lnum,
953 			    int offs)
954 {
955 	struct ubifs_pnode *pnode;
956 	struct ubifs_nbranch *branch;
957 
958 	pnode = pnode_lookup(c, node_num);
959 	if (IS_ERR(pnode))
960 		return PTR_ERR(pnode);
961 	branch = &pnode->parent->nbranch[pnode->iip];
962 	if (branch->lnum != lnum || branch->offs != offs)
963 		return 0;
964 	do_make_pnode_dirty(c, pnode);
965 	return 0;
966 }
967 
968 /**
969  * make_ltab_dirty - make ltab node dirty.
970  * @c: UBIFS file-system description object
971  * @lnum: LEB number where ltab was written
972  * @offs: offset where ltab was written
973  *
974  * This function is used by LPT garbage collection.  LPT garbage collection is
975  * used only for the "big" LPT model (c->big_lpt == 1).  Garbage collection
976  * simply involves marking all the nodes in the LEB being garbage-collected as
977  * dirty.  The dirty nodes are written next commit, after which the LEB is free
978  * to be reused.
979  *
980  * This function returns %0 on success and a negative error code on failure.
981  */
982 static int make_ltab_dirty(struct ubifs_info *c, int lnum, int offs)
983 {
984 	if (lnum != c->ltab_lnum || offs != c->ltab_offs)
985 		return 0; /* This ltab node is obsolete */
986 	if (!(c->lpt_drty_flgs & LTAB_DIRTY)) {
987 		c->lpt_drty_flgs |= LTAB_DIRTY;
988 		ubifs_add_lpt_dirt(c, c->ltab_lnum, c->ltab_sz);
989 	}
990 	return 0;
991 }
992 
993 /**
994  * make_lsave_dirty - make lsave node dirty.
995  * @c: UBIFS file-system description object
996  * @lnum: LEB number where lsave was written
997  * @offs: offset where lsave was written
998  *
999  * This function is used by LPT garbage collection.  LPT garbage collection is
1000  * used only for the "big" LPT model (c->big_lpt == 1).  Garbage collection
1001  * simply involves marking all the nodes in the LEB being garbage-collected as
1002  * dirty.  The dirty nodes are written next commit, after which the LEB is free
1003  * to be reused.
1004  *
1005  * This function returns %0 on success and a negative error code on failure.
1006  */
1007 static int make_lsave_dirty(struct ubifs_info *c, int lnum, int offs)
1008 {
1009 	if (lnum != c->lsave_lnum || offs != c->lsave_offs)
1010 		return 0; /* This lsave node is obsolete */
1011 	if (!(c->lpt_drty_flgs & LSAVE_DIRTY)) {
1012 		c->lpt_drty_flgs |= LSAVE_DIRTY;
1013 		ubifs_add_lpt_dirt(c, c->lsave_lnum, c->lsave_sz);
1014 	}
1015 	return 0;
1016 }
1017 
1018 /**
1019  * make_node_dirty - make node dirty.
1020  * @c: UBIFS file-system description object
1021  * @node_type: LPT node type
1022  * @node_num: node number
1023  * @lnum: LEB number where node was written
1024  * @offs: offset where node was written
1025  *
1026  * This function is used by LPT garbage collection.  LPT garbage collection is
1027  * used only for the "big" LPT model (c->big_lpt == 1).  Garbage collection
1028  * simply involves marking all the nodes in the LEB being garbage-collected as
1029  * dirty.  The dirty nodes are written next commit, after which the LEB is free
1030  * to be reused.
1031  *
1032  * This function returns %0 on success and a negative error code on failure.
1033  */
1034 static int make_node_dirty(struct ubifs_info *c, int node_type, int node_num,
1035 			   int lnum, int offs)
1036 {
1037 	switch (node_type) {
1038 	case UBIFS_LPT_NNODE:
1039 		return make_nnode_dirty(c, node_num, lnum, offs);
1040 	case UBIFS_LPT_PNODE:
1041 		return make_pnode_dirty(c, node_num, lnum, offs);
1042 	case UBIFS_LPT_LTAB:
1043 		return make_ltab_dirty(c, lnum, offs);
1044 	case UBIFS_LPT_LSAVE:
1045 		return make_lsave_dirty(c, lnum, offs);
1046 	}
1047 	return -EINVAL;
1048 }
1049 
1050 /**
1051  * get_lpt_node_len - return the length of a node based on its type.
1052  * @c: UBIFS file-system description object
1053  * @node_type: LPT node type
1054  */
1055 static int get_lpt_node_len(const struct ubifs_info *c, int node_type)
1056 {
1057 	switch (node_type) {
1058 	case UBIFS_LPT_NNODE:
1059 		return c->nnode_sz;
1060 	case UBIFS_LPT_PNODE:
1061 		return c->pnode_sz;
1062 	case UBIFS_LPT_LTAB:
1063 		return c->ltab_sz;
1064 	case UBIFS_LPT_LSAVE:
1065 		return c->lsave_sz;
1066 	}
1067 	return 0;
1068 }
1069 
1070 /**
1071  * get_pad_len - return the length of padding in a buffer.
1072  * @c: UBIFS file-system description object
1073  * @buf: buffer
1074  * @len: length of buffer
1075  */
1076 static int get_pad_len(const struct ubifs_info *c, uint8_t *buf, int len)
1077 {
1078 	int offs, pad_len;
1079 
1080 	if (c->min_io_size == 1)
1081 		return 0;
1082 	offs = c->leb_size - len;
1083 	pad_len = ALIGN(offs, c->min_io_size) - offs;
1084 	return pad_len;
1085 }
1086 
1087 /**
1088  * get_lpt_node_type - return type (and node number) of a node in a buffer.
1089  * @c: UBIFS file-system description object
1090  * @buf: buffer
1091  * @node_num: node number is returned here
1092  */
1093 static int get_lpt_node_type(const struct ubifs_info *c, uint8_t *buf,
1094 			     int *node_num)
1095 {
1096 	uint8_t *addr = buf + UBIFS_LPT_CRC_BYTES;
1097 	int pos = 0, node_type;
1098 
1099 	node_type = ubifs_unpack_bits(&addr, &pos, UBIFS_LPT_TYPE_BITS);
1100 	*node_num = ubifs_unpack_bits(&addr, &pos, c->pcnt_bits);
1101 	return node_type;
1102 }
1103 
1104 /**
1105  * is_a_node - determine if a buffer contains a node.
1106  * @c: UBIFS file-system description object
1107  * @buf: buffer
1108  * @len: length of buffer
1109  *
1110  * This function returns %1 if the buffer contains a node or %0 if it does not.
1111  */
1112 static int is_a_node(const struct ubifs_info *c, uint8_t *buf, int len)
1113 {
1114 	uint8_t *addr = buf + UBIFS_LPT_CRC_BYTES;
1115 	int pos = 0, node_type, node_len;
1116 	uint16_t crc, calc_crc;
1117 
1118 	if (len < UBIFS_LPT_CRC_BYTES + (UBIFS_LPT_TYPE_BITS + 7) / 8)
1119 		return 0;
1120 	node_type = ubifs_unpack_bits(&addr, &pos, UBIFS_LPT_TYPE_BITS);
1121 	if (node_type == UBIFS_LPT_NOT_A_NODE)
1122 		return 0;
1123 	node_len = get_lpt_node_len(c, node_type);
1124 	if (!node_len || node_len > len)
1125 		return 0;
1126 	pos = 0;
1127 	addr = buf;
1128 	crc = ubifs_unpack_bits(&addr, &pos, UBIFS_LPT_CRC_BITS);
1129 	calc_crc = crc16(-1, buf + UBIFS_LPT_CRC_BYTES,
1130 			 node_len - UBIFS_LPT_CRC_BYTES);
1131 	if (crc != calc_crc)
1132 		return 0;
1133 	return 1;
1134 }
1135 
1136 /**
1137  * lpt_gc_lnum - garbage collect a LPT LEB.
1138  * @c: UBIFS file-system description object
1139  * @lnum: LEB number to garbage collect
1140  *
1141  * LPT garbage collection is used only for the "big" LPT model
1142  * (c->big_lpt == 1).  Garbage collection simply involves marking all the nodes
1143  * in the LEB being garbage-collected as dirty.  The dirty nodes are written
1144  * next commit, after which the LEB is free to be reused.
1145  *
1146  * This function returns %0 on success and a negative error code on failure.
1147  */
1148 static int lpt_gc_lnum(struct ubifs_info *c, int lnum)
1149 {
1150 	int err, len = c->leb_size, node_type, node_num, node_len, offs;
1151 	void *buf = c->lpt_buf;
1152 
1153 	dbg_lp("LEB %d", lnum);
1154 
1155 	err = ubifs_leb_read(c, lnum, buf, 0, c->leb_size, 1);
1156 	if (err)
1157 		return err;
1158 
1159 	while (1) {
1160 		if (!is_a_node(c, buf, len)) {
1161 			int pad_len;
1162 
1163 			pad_len = get_pad_len(c, buf, len);
1164 			if (pad_len) {
1165 				buf += pad_len;
1166 				len -= pad_len;
1167 				continue;
1168 			}
1169 			return 0;
1170 		}
1171 		node_type = get_lpt_node_type(c, buf, &node_num);
1172 		node_len = get_lpt_node_len(c, node_type);
1173 		offs = c->leb_size - len;
1174 		ubifs_assert(node_len != 0);
1175 		mutex_lock(&c->lp_mutex);
1176 		err = make_node_dirty(c, node_type, node_num, lnum, offs);
1177 		mutex_unlock(&c->lp_mutex);
1178 		if (err)
1179 			return err;
1180 		buf += node_len;
1181 		len -= node_len;
1182 	}
1183 	return 0;
1184 }
1185 
1186 /**
1187  * lpt_gc - LPT garbage collection.
1188  * @c: UBIFS file-system description object
1189  *
1190  * Select a LPT LEB for LPT garbage collection and call 'lpt_gc_lnum()'.
1191  * Returns %0 on success and a negative error code on failure.
1192  */
1193 static int lpt_gc(struct ubifs_info *c)
1194 {
1195 	int i, lnum = -1, dirty = 0;
1196 
1197 	mutex_lock(&c->lp_mutex);
1198 	for (i = 0; i < c->lpt_lebs; i++) {
1199 		ubifs_assert(!c->ltab[i].tgc);
1200 		if (i + c->lpt_first == c->nhead_lnum ||
1201 		    c->ltab[i].free + c->ltab[i].dirty == c->leb_size)
1202 			continue;
1203 		if (c->ltab[i].dirty > dirty) {
1204 			dirty = c->ltab[i].dirty;
1205 			lnum = i + c->lpt_first;
1206 		}
1207 	}
1208 	mutex_unlock(&c->lp_mutex);
1209 	if (lnum == -1)
1210 		return -ENOSPC;
1211 	return lpt_gc_lnum(c, lnum);
1212 }
1213 
1214 /**
1215  * ubifs_lpt_start_commit - UBIFS commit starts.
1216  * @c: the UBIFS file-system description object
1217  *
1218  * This function has to be called when UBIFS starts the commit operation.
1219  * This function "freezes" all currently dirty LEB properties and does not
1220  * change them anymore. Further changes are saved and tracked separately
1221  * because they are not part of this commit. This function returns zero in case
1222  * of success and a negative error code in case of failure.
1223  */
1224 int ubifs_lpt_start_commit(struct ubifs_info *c)
1225 {
1226 	int err, cnt;
1227 
1228 	dbg_lp("");
1229 
1230 	mutex_lock(&c->lp_mutex);
1231 	err = dbg_chk_lpt_free_spc(c);
1232 	if (err)
1233 		goto out;
1234 	err = dbg_check_ltab(c);
1235 	if (err)
1236 		goto out;
1237 
1238 	if (c->check_lpt_free) {
1239 		/*
1240 		 * We ensure there is enough free space in
1241 		 * ubifs_lpt_post_commit() by marking nodes dirty. That
1242 		 * information is lost when we unmount, so we also need
1243 		 * to check free space once after mounting also.
1244 		 */
1245 		c->check_lpt_free = 0;
1246 		while (need_write_all(c)) {
1247 			mutex_unlock(&c->lp_mutex);
1248 			err = lpt_gc(c);
1249 			if (err)
1250 				return err;
1251 			mutex_lock(&c->lp_mutex);
1252 		}
1253 	}
1254 
1255 	lpt_tgc_start(c);
1256 
1257 	if (!c->dirty_pn_cnt) {
1258 		dbg_cmt("no cnodes to commit");
1259 		err = 0;
1260 		goto out;
1261 	}
1262 
1263 	if (!c->big_lpt && need_write_all(c)) {
1264 		/* If needed, write everything */
1265 		err = make_tree_dirty(c);
1266 		if (err)
1267 			goto out;
1268 		lpt_tgc_start(c);
1269 	}
1270 
1271 	if (c->big_lpt)
1272 		populate_lsave(c);
1273 
1274 	cnt = get_cnodes_to_commit(c);
1275 	ubifs_assert(cnt != 0);
1276 
1277 	err = layout_cnodes(c);
1278 	if (err)
1279 		goto out;
1280 
1281 	/* Copy the LPT's own lprops for end commit to write */
1282 	memcpy(c->ltab_cmt, c->ltab,
1283 	       sizeof(struct ubifs_lpt_lprops) * c->lpt_lebs);
1284 	c->lpt_drty_flgs &= ~(LTAB_DIRTY | LSAVE_DIRTY);
1285 
1286 out:
1287 	mutex_unlock(&c->lp_mutex);
1288 	return err;
1289 }
1290 
1291 /**
1292  * free_obsolete_cnodes - free obsolete cnodes for commit end.
1293  * @c: UBIFS file-system description object
1294  */
1295 static void free_obsolete_cnodes(struct ubifs_info *c)
1296 {
1297 	struct ubifs_cnode *cnode, *cnext;
1298 
1299 	cnext = c->lpt_cnext;
1300 	if (!cnext)
1301 		return;
1302 	do {
1303 		cnode = cnext;
1304 		cnext = cnode->cnext;
1305 		if (test_bit(OBSOLETE_CNODE, &cnode->flags))
1306 			kfree(cnode);
1307 		else
1308 			cnode->cnext = NULL;
1309 	} while (cnext != c->lpt_cnext);
1310 	c->lpt_cnext = NULL;
1311 }
1312 
1313 /**
1314  * ubifs_lpt_end_commit - finish the commit operation.
1315  * @c: the UBIFS file-system description object
1316  *
1317  * This function has to be called when the commit operation finishes. It
1318  * flushes the changes which were "frozen" by 'ubifs_lprops_start_commit()' to
1319  * the media. Returns zero in case of success and a negative error code in case
1320  * of failure.
1321  */
1322 int ubifs_lpt_end_commit(struct ubifs_info *c)
1323 {
1324 	int err;
1325 
1326 	dbg_lp("");
1327 
1328 	if (!c->lpt_cnext)
1329 		return 0;
1330 
1331 	err = write_cnodes(c);
1332 	if (err)
1333 		return err;
1334 
1335 	mutex_lock(&c->lp_mutex);
1336 	free_obsolete_cnodes(c);
1337 	mutex_unlock(&c->lp_mutex);
1338 
1339 	return 0;
1340 }
1341 
1342 /**
1343  * ubifs_lpt_post_commit - post commit LPT trivial GC and LPT GC.
1344  * @c: UBIFS file-system description object
1345  *
1346  * LPT trivial GC is completed after a commit. Also LPT GC is done after a
1347  * commit for the "big" LPT model.
1348  */
1349 int ubifs_lpt_post_commit(struct ubifs_info *c)
1350 {
1351 	int err;
1352 
1353 	mutex_lock(&c->lp_mutex);
1354 	err = lpt_tgc_end(c);
1355 	if (err)
1356 		goto out;
1357 	if (c->big_lpt)
1358 		while (need_write_all(c)) {
1359 			mutex_unlock(&c->lp_mutex);
1360 			err = lpt_gc(c);
1361 			if (err)
1362 				return err;
1363 			mutex_lock(&c->lp_mutex);
1364 		}
1365 out:
1366 	mutex_unlock(&c->lp_mutex);
1367 	return err;
1368 }
1369 
1370 /**
1371  * first_nnode - find the first nnode in memory.
1372  * @c: UBIFS file-system description object
1373  * @hght: height of tree where nnode found is returned here
1374  *
1375  * This function returns a pointer to the nnode found or %NULL if no nnode is
1376  * found. This function is a helper to 'ubifs_lpt_free()'.
1377  */
1378 static struct ubifs_nnode *first_nnode(struct ubifs_info *c, int *hght)
1379 {
1380 	struct ubifs_nnode *nnode;
1381 	int h, i, found;
1382 
1383 	nnode = c->nroot;
1384 	*hght = 0;
1385 	if (!nnode)
1386 		return NULL;
1387 	for (h = 1; h < c->lpt_hght; h++) {
1388 		found = 0;
1389 		for (i = 0; i < UBIFS_LPT_FANOUT; i++) {
1390 			if (nnode->nbranch[i].nnode) {
1391 				found = 1;
1392 				nnode = nnode->nbranch[i].nnode;
1393 				*hght = h;
1394 				break;
1395 			}
1396 		}
1397 		if (!found)
1398 			break;
1399 	}
1400 	return nnode;
1401 }
1402 
1403 /**
1404  * next_nnode - find the next nnode in memory.
1405  * @c: UBIFS file-system description object
1406  * @nnode: nnode from which to start.
1407  * @hght: height of tree where nnode is, is passed and returned here
1408  *
1409  * This function returns a pointer to the nnode found or %NULL if no nnode is
1410  * found. This function is a helper to 'ubifs_lpt_free()'.
1411  */
1412 static struct ubifs_nnode *next_nnode(struct ubifs_info *c,
1413 				      struct ubifs_nnode *nnode, int *hght)
1414 {
1415 	struct ubifs_nnode *parent;
1416 	int iip, h, i, found;
1417 
1418 	parent = nnode->parent;
1419 	if (!parent)
1420 		return NULL;
1421 	if (nnode->iip == UBIFS_LPT_FANOUT - 1) {
1422 		*hght -= 1;
1423 		return parent;
1424 	}
1425 	for (iip = nnode->iip + 1; iip < UBIFS_LPT_FANOUT; iip++) {
1426 		nnode = parent->nbranch[iip].nnode;
1427 		if (nnode)
1428 			break;
1429 	}
1430 	if (!nnode) {
1431 		*hght -= 1;
1432 		return parent;
1433 	}
1434 	for (h = *hght + 1; h < c->lpt_hght; h++) {
1435 		found = 0;
1436 		for (i = 0; i < UBIFS_LPT_FANOUT; i++) {
1437 			if (nnode->nbranch[i].nnode) {
1438 				found = 1;
1439 				nnode = nnode->nbranch[i].nnode;
1440 				*hght = h;
1441 				break;
1442 			}
1443 		}
1444 		if (!found)
1445 			break;
1446 	}
1447 	return nnode;
1448 }
1449 
1450 /**
1451  * ubifs_lpt_free - free resources owned by the LPT.
1452  * @c: UBIFS file-system description object
1453  * @wr_only: free only resources used for writing
1454  */
1455 void ubifs_lpt_free(struct ubifs_info *c, int wr_only)
1456 {
1457 	struct ubifs_nnode *nnode;
1458 	int i, hght;
1459 
1460 	/* Free write-only things first */
1461 
1462 	free_obsolete_cnodes(c); /* Leftover from a failed commit */
1463 
1464 	vfree(c->ltab_cmt);
1465 	c->ltab_cmt = NULL;
1466 	vfree(c->lpt_buf);
1467 	c->lpt_buf = NULL;
1468 	kfree(c->lsave);
1469 	c->lsave = NULL;
1470 
1471 	if (wr_only)
1472 		return;
1473 
1474 	/* Now free the rest */
1475 
1476 	nnode = first_nnode(c, &hght);
1477 	while (nnode) {
1478 		for (i = 0; i < UBIFS_LPT_FANOUT; i++)
1479 			kfree(nnode->nbranch[i].nnode);
1480 		nnode = next_nnode(c, nnode, &hght);
1481 	}
1482 	for (i = 0; i < LPROPS_HEAP_CNT; i++)
1483 		kfree(c->lpt_heap[i].arr);
1484 	kfree(c->dirty_idx.arr);
1485 	kfree(c->nroot);
1486 	vfree(c->ltab);
1487 	kfree(c->lpt_nod_buf);
1488 }
1489 
1490 /*
1491  * Everything below is related to debugging.
1492  */
1493 
1494 /**
1495  * dbg_is_all_ff - determine if a buffer contains only 0xFF bytes.
1496  * @buf: buffer
1497  * @len: buffer length
1498  */
1499 static int dbg_is_all_ff(uint8_t *buf, int len)
1500 {
1501 	int i;
1502 
1503 	for (i = 0; i < len; i++)
1504 		if (buf[i] != 0xff)
1505 			return 0;
1506 	return 1;
1507 }
1508 
1509 /**
1510  * dbg_is_nnode_dirty - determine if a nnode is dirty.
1511  * @c: the UBIFS file-system description object
1512  * @lnum: LEB number where nnode was written
1513  * @offs: offset where nnode was written
1514  */
1515 static int dbg_is_nnode_dirty(struct ubifs_info *c, int lnum, int offs)
1516 {
1517 	struct ubifs_nnode *nnode;
1518 	int hght;
1519 
1520 	/* Entire tree is in memory so first_nnode / next_nnode are OK */
1521 	nnode = first_nnode(c, &hght);
1522 	for (; nnode; nnode = next_nnode(c, nnode, &hght)) {
1523 		struct ubifs_nbranch *branch;
1524 
1525 		cond_resched();
1526 		if (nnode->parent) {
1527 			branch = &nnode->parent->nbranch[nnode->iip];
1528 			if (branch->lnum != lnum || branch->offs != offs)
1529 				continue;
1530 			if (test_bit(DIRTY_CNODE, &nnode->flags))
1531 				return 1;
1532 			return 0;
1533 		} else {
1534 			if (c->lpt_lnum != lnum || c->lpt_offs != offs)
1535 				continue;
1536 			if (test_bit(DIRTY_CNODE, &nnode->flags))
1537 				return 1;
1538 			return 0;
1539 		}
1540 	}
1541 	return 1;
1542 }
1543 
1544 /**
1545  * dbg_is_pnode_dirty - determine if a pnode is dirty.
1546  * @c: the UBIFS file-system description object
1547  * @lnum: LEB number where pnode was written
1548  * @offs: offset where pnode was written
1549  */
1550 static int dbg_is_pnode_dirty(struct ubifs_info *c, int lnum, int offs)
1551 {
1552 	int i, cnt;
1553 
1554 	cnt = DIV_ROUND_UP(c->main_lebs, UBIFS_LPT_FANOUT);
1555 	for (i = 0; i < cnt; i++) {
1556 		struct ubifs_pnode *pnode;
1557 		struct ubifs_nbranch *branch;
1558 
1559 		cond_resched();
1560 		pnode = pnode_lookup(c, i);
1561 		if (IS_ERR(pnode))
1562 			return PTR_ERR(pnode);
1563 		branch = &pnode->parent->nbranch[pnode->iip];
1564 		if (branch->lnum != lnum || branch->offs != offs)
1565 			continue;
1566 		if (test_bit(DIRTY_CNODE, &pnode->flags))
1567 			return 1;
1568 		return 0;
1569 	}
1570 	return 1;
1571 }
1572 
1573 /**
1574  * dbg_is_ltab_dirty - determine if a ltab node is dirty.
1575  * @c: the UBIFS file-system description object
1576  * @lnum: LEB number where ltab node was written
1577  * @offs: offset where ltab node was written
1578  */
1579 static int dbg_is_ltab_dirty(struct ubifs_info *c, int lnum, int offs)
1580 {
1581 	if (lnum != c->ltab_lnum || offs != c->ltab_offs)
1582 		return 1;
1583 	return (c->lpt_drty_flgs & LTAB_DIRTY) != 0;
1584 }
1585 
1586 /**
1587  * dbg_is_lsave_dirty - determine if a lsave node is dirty.
1588  * @c: the UBIFS file-system description object
1589  * @lnum: LEB number where lsave node was written
1590  * @offs: offset where lsave node was written
1591  */
1592 static int dbg_is_lsave_dirty(struct ubifs_info *c, int lnum, int offs)
1593 {
1594 	if (lnum != c->lsave_lnum || offs != c->lsave_offs)
1595 		return 1;
1596 	return (c->lpt_drty_flgs & LSAVE_DIRTY) != 0;
1597 }
1598 
1599 /**
1600  * dbg_is_node_dirty - determine if a node is dirty.
1601  * @c: the UBIFS file-system description object
1602  * @node_type: node type
1603  * @lnum: LEB number where node was written
1604  * @offs: offset where node was written
1605  */
1606 static int dbg_is_node_dirty(struct ubifs_info *c, int node_type, int lnum,
1607 			     int offs)
1608 {
1609 	switch (node_type) {
1610 	case UBIFS_LPT_NNODE:
1611 		return dbg_is_nnode_dirty(c, lnum, offs);
1612 	case UBIFS_LPT_PNODE:
1613 		return dbg_is_pnode_dirty(c, lnum, offs);
1614 	case UBIFS_LPT_LTAB:
1615 		return dbg_is_ltab_dirty(c, lnum, offs);
1616 	case UBIFS_LPT_LSAVE:
1617 		return dbg_is_lsave_dirty(c, lnum, offs);
1618 	}
1619 	return 1;
1620 }
1621 
1622 /**
1623  * dbg_check_ltab_lnum - check the ltab for a LPT LEB number.
1624  * @c: the UBIFS file-system description object
1625  * @lnum: LEB number where node was written
1626  * @offs: offset where node was written
1627  *
1628  * This function returns %0 on success and a negative error code on failure.
1629  */
1630 static int dbg_check_ltab_lnum(struct ubifs_info *c, int lnum)
1631 {
1632 	int err, len = c->leb_size, dirty = 0, node_type, node_num, node_len;
1633 	int ret;
1634 	void *buf, *p;
1635 
1636 	if (!dbg_is_chk_lprops(c))
1637 		return 0;
1638 
1639 	buf = p = __vmalloc(c->leb_size, GFP_NOFS, PAGE_KERNEL);
1640 	if (!buf) {
1641 		ubifs_err(c, "cannot allocate memory for ltab checking");
1642 		return 0;
1643 	}
1644 
1645 	dbg_lp("LEB %d", lnum);
1646 
1647 	err = ubifs_leb_read(c, lnum, buf, 0, c->leb_size, 1);
1648 	if (err)
1649 		goto out;
1650 
1651 	while (1) {
1652 		if (!is_a_node(c, p, len)) {
1653 			int i, pad_len;
1654 
1655 			pad_len = get_pad_len(c, p, len);
1656 			if (pad_len) {
1657 				p += pad_len;
1658 				len -= pad_len;
1659 				dirty += pad_len;
1660 				continue;
1661 			}
1662 			if (!dbg_is_all_ff(p, len)) {
1663 				ubifs_err(c, "invalid empty space in LEB %d at %d",
1664 					  lnum, c->leb_size - len);
1665 				err = -EINVAL;
1666 			}
1667 			i = lnum - c->lpt_first;
1668 			if (len != c->ltab[i].free) {
1669 				ubifs_err(c, "invalid free space in LEB %d (free %d, expected %d)",
1670 					  lnum, len, c->ltab[i].free);
1671 				err = -EINVAL;
1672 			}
1673 			if (dirty != c->ltab[i].dirty) {
1674 				ubifs_err(c, "invalid dirty space in LEB %d (dirty %d, expected %d)",
1675 					  lnum, dirty, c->ltab[i].dirty);
1676 				err = -EINVAL;
1677 			}
1678 			goto out;
1679 		}
1680 		node_type = get_lpt_node_type(c, p, &node_num);
1681 		node_len = get_lpt_node_len(c, node_type);
1682 		ret = dbg_is_node_dirty(c, node_type, lnum, c->leb_size - len);
1683 		if (ret == 1)
1684 			dirty += node_len;
1685 		p += node_len;
1686 		len -= node_len;
1687 	}
1688 
1689 	err = 0;
1690 out:
1691 	vfree(buf);
1692 	return err;
1693 }
1694 
1695 /**
1696  * dbg_check_ltab - check the free and dirty space in the ltab.
1697  * @c: the UBIFS file-system description object
1698  *
1699  * This function returns %0 on success and a negative error code on failure.
1700  */
1701 int dbg_check_ltab(struct ubifs_info *c)
1702 {
1703 	int lnum, err, i, cnt;
1704 
1705 	if (!dbg_is_chk_lprops(c))
1706 		return 0;
1707 
1708 	/* Bring the entire tree into memory */
1709 	cnt = DIV_ROUND_UP(c->main_lebs, UBIFS_LPT_FANOUT);
1710 	for (i = 0; i < cnt; i++) {
1711 		struct ubifs_pnode *pnode;
1712 
1713 		pnode = pnode_lookup(c, i);
1714 		if (IS_ERR(pnode))
1715 			return PTR_ERR(pnode);
1716 		cond_resched();
1717 	}
1718 
1719 	/* Check nodes */
1720 	err = dbg_check_lpt_nodes(c, (struct ubifs_cnode *)c->nroot, 0, 0);
1721 	if (err)
1722 		return err;
1723 
1724 	/* Check each LEB */
1725 	for (lnum = c->lpt_first; lnum <= c->lpt_last; lnum++) {
1726 		err = dbg_check_ltab_lnum(c, lnum);
1727 		if (err) {
1728 			ubifs_err(c, "failed at LEB %d", lnum);
1729 			return err;
1730 		}
1731 	}
1732 
1733 	dbg_lp("succeeded");
1734 	return 0;
1735 }
1736 
1737 /**
1738  * dbg_chk_lpt_free_spc - check LPT free space is enough to write entire LPT.
1739  * @c: the UBIFS file-system description object
1740  *
1741  * This function returns %0 on success and a negative error code on failure.
1742  */
1743 int dbg_chk_lpt_free_spc(struct ubifs_info *c)
1744 {
1745 	long long free = 0;
1746 	int i;
1747 
1748 	if (!dbg_is_chk_lprops(c))
1749 		return 0;
1750 
1751 	for (i = 0; i < c->lpt_lebs; i++) {
1752 		if (c->ltab[i].tgc || c->ltab[i].cmt)
1753 			continue;
1754 		if (i + c->lpt_first == c->nhead_lnum)
1755 			free += c->leb_size - c->nhead_offs;
1756 		else if (c->ltab[i].free == c->leb_size)
1757 			free += c->leb_size;
1758 	}
1759 	if (free < c->lpt_sz) {
1760 		ubifs_err(c, "LPT space error: free %lld lpt_sz %lld",
1761 			  free, c->lpt_sz);
1762 		ubifs_dump_lpt_info(c);
1763 		ubifs_dump_lpt_lebs(c);
1764 		dump_stack();
1765 		return -EINVAL;
1766 	}
1767 	return 0;
1768 }
1769 
1770 /**
1771  * dbg_chk_lpt_sz - check LPT does not write more than LPT size.
1772  * @c: the UBIFS file-system description object
1773  * @action: what to do
1774  * @len: length written
1775  *
1776  * This function returns %0 on success and a negative error code on failure.
1777  * The @action argument may be one of:
1778  *   o %0 - LPT debugging checking starts, initialize debugging variables;
1779  *   o %1 - wrote an LPT node, increase LPT size by @len bytes;
1780  *   o %2 - switched to a different LEB and wasted @len bytes;
1781  *   o %3 - check that we've written the right number of bytes.
1782  *   o %4 - wasted @len bytes;
1783  */
1784 int dbg_chk_lpt_sz(struct ubifs_info *c, int action, int len)
1785 {
1786 	struct ubifs_debug_info *d = c->dbg;
1787 	long long chk_lpt_sz, lpt_sz;
1788 	int err = 0;
1789 
1790 	if (!dbg_is_chk_lprops(c))
1791 		return 0;
1792 
1793 	switch (action) {
1794 	case 0:
1795 		d->chk_lpt_sz = 0;
1796 		d->chk_lpt_sz2 = 0;
1797 		d->chk_lpt_lebs = 0;
1798 		d->chk_lpt_wastage = 0;
1799 		if (c->dirty_pn_cnt > c->pnode_cnt) {
1800 			ubifs_err(c, "dirty pnodes %d exceed max %d",
1801 				  c->dirty_pn_cnt, c->pnode_cnt);
1802 			err = -EINVAL;
1803 		}
1804 		if (c->dirty_nn_cnt > c->nnode_cnt) {
1805 			ubifs_err(c, "dirty nnodes %d exceed max %d",
1806 				  c->dirty_nn_cnt, c->nnode_cnt);
1807 			err = -EINVAL;
1808 		}
1809 		return err;
1810 	case 1:
1811 		d->chk_lpt_sz += len;
1812 		return 0;
1813 	case 2:
1814 		d->chk_lpt_sz += len;
1815 		d->chk_lpt_wastage += len;
1816 		d->chk_lpt_lebs += 1;
1817 		return 0;
1818 	case 3:
1819 		chk_lpt_sz = c->leb_size;
1820 		chk_lpt_sz *= d->chk_lpt_lebs;
1821 		chk_lpt_sz += len - c->nhead_offs;
1822 		if (d->chk_lpt_sz != chk_lpt_sz) {
1823 			ubifs_err(c, "LPT wrote %lld but space used was %lld",
1824 				  d->chk_lpt_sz, chk_lpt_sz);
1825 			err = -EINVAL;
1826 		}
1827 		if (d->chk_lpt_sz > c->lpt_sz) {
1828 			ubifs_err(c, "LPT wrote %lld but lpt_sz is %lld",
1829 				  d->chk_lpt_sz, c->lpt_sz);
1830 			err = -EINVAL;
1831 		}
1832 		if (d->chk_lpt_sz2 && d->chk_lpt_sz != d->chk_lpt_sz2) {
1833 			ubifs_err(c, "LPT layout size %lld but wrote %lld",
1834 				  d->chk_lpt_sz, d->chk_lpt_sz2);
1835 			err = -EINVAL;
1836 		}
1837 		if (d->chk_lpt_sz2 && d->new_nhead_offs != len) {
1838 			ubifs_err(c, "LPT new nhead offs: expected %d was %d",
1839 				  d->new_nhead_offs, len);
1840 			err = -EINVAL;
1841 		}
1842 		lpt_sz = (long long)c->pnode_cnt * c->pnode_sz;
1843 		lpt_sz += (long long)c->nnode_cnt * c->nnode_sz;
1844 		lpt_sz += c->ltab_sz;
1845 		if (c->big_lpt)
1846 			lpt_sz += c->lsave_sz;
1847 		if (d->chk_lpt_sz - d->chk_lpt_wastage > lpt_sz) {
1848 			ubifs_err(c, "LPT chk_lpt_sz %lld + waste %lld exceeds %lld",
1849 				  d->chk_lpt_sz, d->chk_lpt_wastage, lpt_sz);
1850 			err = -EINVAL;
1851 		}
1852 		if (err) {
1853 			ubifs_dump_lpt_info(c);
1854 			ubifs_dump_lpt_lebs(c);
1855 			dump_stack();
1856 		}
1857 		d->chk_lpt_sz2 = d->chk_lpt_sz;
1858 		d->chk_lpt_sz = 0;
1859 		d->chk_lpt_wastage = 0;
1860 		d->chk_lpt_lebs = 0;
1861 		d->new_nhead_offs = len;
1862 		return err;
1863 	case 4:
1864 		d->chk_lpt_sz += len;
1865 		d->chk_lpt_wastage += len;
1866 		return 0;
1867 	default:
1868 		return -EINVAL;
1869 	}
1870 }
1871 
1872 /**
1873  * ubifs_dump_lpt_leb - dump an LPT LEB.
1874  * @c: UBIFS file-system description object
1875  * @lnum: LEB number to dump
1876  *
1877  * This function dumps an LEB from LPT area. Nodes in this area are very
1878  * different to nodes in the main area (e.g., they do not have common headers,
1879  * they do not have 8-byte alignments, etc), so we have a separate function to
1880  * dump LPT area LEBs. Note, LPT has to be locked by the caller.
1881  */
1882 static void dump_lpt_leb(const struct ubifs_info *c, int lnum)
1883 {
1884 	int err, len = c->leb_size, node_type, node_num, node_len, offs;
1885 	void *buf, *p;
1886 
1887 	pr_err("(pid %d) start dumping LEB %d\n", current->pid, lnum);
1888 	buf = p = __vmalloc(c->leb_size, GFP_NOFS, PAGE_KERNEL);
1889 	if (!buf) {
1890 		ubifs_err(c, "cannot allocate memory to dump LPT");
1891 		return;
1892 	}
1893 
1894 	err = ubifs_leb_read(c, lnum, buf, 0, c->leb_size, 1);
1895 	if (err)
1896 		goto out;
1897 
1898 	while (1) {
1899 		offs = c->leb_size - len;
1900 		if (!is_a_node(c, p, len)) {
1901 			int pad_len;
1902 
1903 			pad_len = get_pad_len(c, p, len);
1904 			if (pad_len) {
1905 				pr_err("LEB %d:%d, pad %d bytes\n",
1906 				       lnum, offs, pad_len);
1907 				p += pad_len;
1908 				len -= pad_len;
1909 				continue;
1910 			}
1911 			if (len)
1912 				pr_err("LEB %d:%d, free %d bytes\n",
1913 				       lnum, offs, len);
1914 			break;
1915 		}
1916 
1917 		node_type = get_lpt_node_type(c, p, &node_num);
1918 		switch (node_type) {
1919 		case UBIFS_LPT_PNODE:
1920 		{
1921 			node_len = c->pnode_sz;
1922 			if (c->big_lpt)
1923 				pr_err("LEB %d:%d, pnode num %d\n",
1924 				       lnum, offs, node_num);
1925 			else
1926 				pr_err("LEB %d:%d, pnode\n", lnum, offs);
1927 			break;
1928 		}
1929 		case UBIFS_LPT_NNODE:
1930 		{
1931 			int i;
1932 			struct ubifs_nnode nnode;
1933 
1934 			node_len = c->nnode_sz;
1935 			if (c->big_lpt)
1936 				pr_err("LEB %d:%d, nnode num %d, ",
1937 				       lnum, offs, node_num);
1938 			else
1939 				pr_err("LEB %d:%d, nnode, ",
1940 				       lnum, offs);
1941 			err = ubifs_unpack_nnode(c, p, &nnode);
1942 			if (err) {
1943 				pr_err("failed to unpack_node, error %d\n",
1944 				       err);
1945 				break;
1946 			}
1947 			for (i = 0; i < UBIFS_LPT_FANOUT; i++) {
1948 				pr_cont("%d:%d", nnode.nbranch[i].lnum,
1949 				       nnode.nbranch[i].offs);
1950 				if (i != UBIFS_LPT_FANOUT - 1)
1951 					pr_cont(", ");
1952 			}
1953 			pr_cont("\n");
1954 			break;
1955 		}
1956 		case UBIFS_LPT_LTAB:
1957 			node_len = c->ltab_sz;
1958 			pr_err("LEB %d:%d, ltab\n", lnum, offs);
1959 			break;
1960 		case UBIFS_LPT_LSAVE:
1961 			node_len = c->lsave_sz;
1962 			pr_err("LEB %d:%d, lsave len\n", lnum, offs);
1963 			break;
1964 		default:
1965 			ubifs_err(c, "LPT node type %d not recognized", node_type);
1966 			goto out;
1967 		}
1968 
1969 		p += node_len;
1970 		len -= node_len;
1971 	}
1972 
1973 	pr_err("(pid %d) finish dumping LEB %d\n", current->pid, lnum);
1974 out:
1975 	vfree(buf);
1976 	return;
1977 }
1978 
1979 /**
1980  * ubifs_dump_lpt_lebs - dump LPT lebs.
1981  * @c: UBIFS file-system description object
1982  *
1983  * This function dumps all LPT LEBs. The caller has to make sure the LPT is
1984  * locked.
1985  */
1986 void ubifs_dump_lpt_lebs(const struct ubifs_info *c)
1987 {
1988 	int i;
1989 
1990 	pr_err("(pid %d) start dumping all LPT LEBs\n", current->pid);
1991 	for (i = 0; i < c->lpt_lebs; i++)
1992 		dump_lpt_leb(c, i + c->lpt_first);
1993 	pr_err("(pid %d) finish dumping all LPT LEBs\n", current->pid);
1994 }
1995 
1996 /**
1997  * dbg_populate_lsave - debugging version of 'populate_lsave()'
1998  * @c: UBIFS file-system description object
1999  *
2000  * This is a debugging version for 'populate_lsave()' which populates lsave
2001  * with random LEBs instead of useful LEBs, which is good for test coverage.
2002  * Returns zero if lsave has not been populated (this debugging feature is
2003  * disabled) an non-zero if lsave has been populated.
2004  */
2005 static int dbg_populate_lsave(struct ubifs_info *c)
2006 {
2007 	struct ubifs_lprops *lprops;
2008 	struct ubifs_lpt_heap *heap;
2009 	int i;
2010 
2011 	if (!dbg_is_chk_gen(c))
2012 		return 0;
2013 	if (prandom_u32() & 3)
2014 		return 0;
2015 
2016 	for (i = 0; i < c->lsave_cnt; i++)
2017 		c->lsave[i] = c->main_first;
2018 
2019 	list_for_each_entry(lprops, &c->empty_list, list)
2020 		c->lsave[prandom_u32() % c->lsave_cnt] = lprops->lnum;
2021 	list_for_each_entry(lprops, &c->freeable_list, list)
2022 		c->lsave[prandom_u32() % c->lsave_cnt] = lprops->lnum;
2023 	list_for_each_entry(lprops, &c->frdi_idx_list, list)
2024 		c->lsave[prandom_u32() % c->lsave_cnt] = lprops->lnum;
2025 
2026 	heap = &c->lpt_heap[LPROPS_DIRTY_IDX - 1];
2027 	for (i = 0; i < heap->cnt; i++)
2028 		c->lsave[prandom_u32() % c->lsave_cnt] = heap->arr[i]->lnum;
2029 	heap = &c->lpt_heap[LPROPS_DIRTY - 1];
2030 	for (i = 0; i < heap->cnt; i++)
2031 		c->lsave[prandom_u32() % c->lsave_cnt] = heap->arr[i]->lnum;
2032 	heap = &c->lpt_heap[LPROPS_FREE - 1];
2033 	for (i = 0; i < heap->cnt; i++)
2034 		c->lsave[prandom_u32() % c->lsave_cnt] = heap->arr[i]->lnum;
2035 
2036 	return 1;
2037 }
2038